Face method-a figurative representation of multivariate data-originally developed by H. Chernoff has been applied in a lot of fields. However, face method has the difficulty of efficiently assigning multivariate data components to respective facial features of face pattern(variables assignment), and the efficiency of the assignment determines the accuracy of the results of face method. For the moment owing to the absence of a developed method, the assignment is generally based on heuristic trial. In this paper, when certain multivariate data are known to have involved some groups, a search algorithm for an assignment method is proposed to separate psychometrical distances among facial expressions represented by the data of some groups, that is, to classify the differences among the groups. The algorithm is applied to the fossil data of Chernoff's study and its efficiency is tested.